OpenAI

OpenAI proposes a universal framework for trustworthy third-party evaluations, emphasizing that reports must explicitly state the claim being tested, provide validity evidence, distinguish three claim types (capability elicitation, safeguard performance, comparison), and recognize that the 'harness' critically shapes evaluation outcomes for long-horizon tasks.

The AI industry is advancing the 'dead economy theory' through hundreds of billions in investment: its true goal is wholesale replacement of the global labor market—not augmentation. Current valuations depend on large-scale human cost elimination; otherwise, they represent capitalism’s largest bubble.

GPT 5.5 and other models' capability improvements are not sudden jumps but result of model reliability crossing a key threshold. Reinforcement learning, post-training optimization, and evolving evaluation systems drive AI practicality.

Google invested $40 billion in Anthropic's secret AI model, signaling strategic importance.

Virgin Atlantic used Codex to launch its new mobile app before the high-risk Christmas rush, achieving zero P1 defects and near-full test coverage, while accelerating legacy refactoring by up to 80% and enabling non-engineers to build data apps in hours.

The TeamPCP group has launched over 20 waves of supply chain attacks in months, compromising more than 500 distinct open-source tools—over 1,000 versions—and exposing at least 3,800 internal GitHub repositories.

An internal OpenAI model disproved the famous Erdős unit distance conjecture, demonstrating AI's breakthrough capability in mathematical research.

本文介绍了如何通过Auth Proxy来保护LangSmith代理沙箱的网络访问，确保在大规模部署代理时的安全性。Auth Proxy通过在网络层控制和管理代理与外部服务的交互，实现了凭据的安全管理、网络访问的显式控制以及团队职责的清晰分离。

OpenAI built a sandbox for Codex on Windows using dual local users and restricted tokens, solving the lack of native process isolation and enabling secure default execution.

GPT-2 demonstrated that training a large language model solely on unsupervised next-word prediction enables emergent multitask capabilities, performing translation, QA, and summarization without task-specific fine-tuning.

OpenAI introduces three real-time audio models to enhance voice applications.

OpenAI, in collaboration with AMD, NVIDIA, and others, introduces MRC—a new networking protocol that enhances performance and reliability for large-scale AI training, now open-sourced via OCP to advance industry standards.

Qwen's new desktop voice input enables global activation, mixed Chinese-English recognition, and AI-powered content generation and task execution, enabling truly hands-free, efficient office work.

OpenAI, together with AMD, Intel, NVIDIA and others, has open-sourced MRC, a new networking protocol that improves reliability and efficiency in large-scale AI training clusters.

A 15-person Chinese team, Luma AI, launched Uni-1.1, an AI image model that integrates reasoning and generation, slashes costs by 50%, and achieves top-3 global ranking on Arena.ai—offering the most controllable, scalable solution for brand visual production beyond OpenAI and Google.

OpenAI reveals its rearchitected WebRTC stack that decouples media termination from routing using a relay+transceiver model, enabling global low-latency voice AI for 900M+ users by solving ICE/DTLS statefulness and Kubernetes deployment conflicts.

OpenAI Codex 推出 Auto-review 模式：用独立 AI Agent 替代人工审批越界行为，在安全与可用性间实现新平衡，自动批准率超99%，打扰人类频率降低200倍。

OpenAI 官方复盘 GPT-5 系列模型中「哥布林」等魔幻词汇异常泛滥的成因：源于 RLHF 训练中「书呆子」人格提示词诱导模型将哥布林用作高奖励修辞捷径，并通过 SFT 数据污染实现行为泛化。

OpenAI's AI model has autonomously disproved a central conjecture in discrete geometry, demonstrating polynomially improved unit-distance point set constructions, marking a milestone in AI-driven mathematical research.

OpenAI built a 131,000 GPU training network using the MRC protocol with counterintuitive design choices that significantly improve training efficiency.

OpenAI launches Daybreak, a strategic AI-powered cybersecurity initiative that embeds security into software design from the start.

Stripe在年度大会发布288项新功能，旨在构建AI时代的经济基础设施，包括支持智能体的商务套件、Link钱包、Streaming Payments、强化风控及Treasury扩展等，助力企业适应AI变革。

OpenAI released three new voice models: GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper, significantly enhancing dialogue, translation, and real-time transcription capabilities.

Crossover Intelligence topped the World Arena Track 2 with its DSCFuncWorld, outpacing the second-place model by a significant margin and validating end-to-end data generation, strategy training, and task execution capabilities.

Uber caps AI tool spending per month at $1,500 per tool, independent of other tools, and applies only to agentic coding software such as Cursor or Claude Code. At $3,000/year per engineer for two tools, the cap represents roughly 11% of the median $330,000 software engineer salary in the U.S.

Assembly AI shipped major advancements in May: LLM Gateway now supports reasoning via a single parameter (low/medium/high) and abstracts provider differences; streaming speaker diarization accuracy jumped with 66% fewer false alarms and 60% fewer phantom turns, per-word speaker labels, and continuous partials every ~3 seconds; Playground now previews 34 voices and allows public sharing; PII redaction is live for streaming, automatically removing names, phone numbers, credit card numbers in real

NVIDIA introduces the 550B-parameter Neotron 3 Ultra, a mixture-of-experts agent model trained for task orchestration, outperforming many trillion-parameter open agents on benchmarks, with full data and recipe transparency to enable enterprise on-prem deployment and fine-tuning.

As AI matured, Stripe France pivoted to platform- and brand-centric subscription models, integrating OpenAI first on Monde. The result: mentions in GPT drive 20x traffic vs. MTA articles and 50x vs. Discover/Google pieces.

Advances in image composition are simultaneously broken by Reve 2 and Ideogram 4, with Ideogram 4 now the top-ranked open image model on Arena. Microsoft released MAI-Thinking-1 achieving 97% on AIME 2025 without synthetic data or distillation, publishing detailed training stacks and MoE scaling. Frontier Tuning enables enterprise workflow models to reach GPT-5.4 quality with up to 10× efficiency gains, while Gemma 4 12B and others strengthen local-first deployment momentum.

Wasmer built Edge.js in two weeks using OpenAI Codex, enabling Node.js workloads to run safely inside a WebAssembly sandbox without Docker. Development speed increased 10–20x, and it became the first cloud host to provide full Node.js at the edge.